Qdrant — Vector Similarity Search

High-performance vector database written in Rust, designed for production RAG and semantic search workloads. Provides hybrid search (vectors + metadata filtering), multi-vector support, quantization for memory efficiency, and distributed operation via Raft with sharding and replication. Includes REST and gRPC APIs and first-class Python client.

When to use

• Production RAG systems requiring low-latency nearest-neighbor search
• Use cases needing metadata filtering alongside vector similarity
• Large-scale collections where quantization or sharding is required

Key features

• Rust-powered performance and safety
• HNSW indexing and multiple distance metrics (COSINE, EUCLID, DOT, MANHATTAN)
• Payload indexing and rich filter support
• Multi/vector (named vectors), sparse vectors, and on-disk payloads
• Quantization and tuning knobs for memory vs accuracy tradeoffs

Quick start (high level)

• Install qdrant or run via Docker; pip install qdrant-client for Python
• Create collections with VectorParams and desired distance metric
• Batch upsert points (id + vector + payload) and build payload indexes for filters
• Use client.search or search_batch for retrieval and RAG pipelines

Integrations

• Works with sentence-transformers, LangChain, LlamaIndex and common embedding pipelines

References

• Qdrant docs: https://qdrant.tech/documentation/
• GitHub: https://github.com/qdrant/qdrant